Recite

AI-powered receipt scanning that renames files by date/vendor, extracts transaction details, and logs them in a dynamic bookkeeping CSV.

Before installing or running this skill: (1) Understand that receipts (full images/PDFs) are uploaded to a third-party endpoint (recite.rivra.dev). Only proceed if you trust that service and its privacy/security practices. (2) The registry metadata does not list the required RECITE_API_KEY even though both SKILL.md and the script require it — treat the key as mandatory. (3) The SKILL.md promises behavior (reading long_term_memory.md and skipping rows when fields disappear) that the script does not implement; expect the script to write blanks or add new columns rather than skipping entries. (4) Back up your receipt folder before a run (files will be renamed) and test on a small set first. (5) If you need stronger privacy guarantees, request or implement an offline/local OCR alternative instead of sending images to an external API. (6) If you plan to use this skill in production, ask the author for corrected registry metadata, documentation that matches the code, and a privacy/security statement for the recite.rivra.dev endpoint.

Type: OpenClaw Skill Name: recite Version: 1.0.1 The skill's core functionality is benign, but the `process_receipts.py` script contains a path traversal vulnerability. It uses `sys.argv[1]` (target directory) and API-provided `vendor` and `date` fields directly in `os.path.join` and `os.rename` without robust sanitization, potentially allowing files to be renamed or moved to arbitrary locations if malicious input or API responses are provided. Furthermore, `SKILL.md` and `README.md` instruct the AI agent to read `long_term_memory.md` for custom instructions, creating a prompt injection surface against the agent, though the skill itself does not contain malicious instructions in this file.